Abstract
<jats:p>Verifying the claimed harvest location of timber is critical for enforcement of national and international trade regulations. One approach is to develop harvest-location models that compare the stable isotope ratio profile of a test wood sample with maps of predicted isotope ratios, known as isoscapes. Here, we test the power and suitability of a Bayesian framework for inferring the harvest location of timber in simulated isoscapes and real-world reference data from the Eastern United States, Eastern Europe, and Peru. We systematically vary reference dataset size, sampling strategy, isotope choice, isoscape residual variance, and isoscape autocorrelation range to evaluate how these parameters affect model performance. We find that reference dataset sample count and isotope choice are both important controls on the accuracy of the inferred harvest location. Sample count and isotope choice also colimit inversion accuracy, such that increasing both axes of reference dataset size is necessary to improve the accuracy of the inferred harvest location. Furthermore, reference datasets that are randomly distributed throughout a study area outperform reference datasets that repeatedly sample near the same locations of a study area. Isotope ratios that exhibit subtle, small-scale variability are the most difficult to model accurately, even when measurement error is ignored and when simulating a large reference dataset. Yet, including these challenging isotope ratios in our experiments with real-world data still improves the accuracy of the inferred harvest location. To improve model performance, we recommend that practitioners prioritize reference dataset size over spatial sampling design and investigate the spatial scale at which stable isotope ratios vary in their reference data.</jats:p>